Irtorgb - Revision history

imported>Student2018: /* CNN model result */

2018-12-17T00:48:04Z

CNN model result

← Older revision		Revision as of 00:48, 17 December 2018
Line 271:		Line 271:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Train trueline.png\|250px\|thumb\|left\|alt=A\|'''Fig. 18. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Train trueline.png\|250px\|thumb\|left\|alt=A\|'''Fig. 18. ''' From up to dowm: R, G, B. (blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line)]]
	\| [[File:Train result.png\|650px\|thumb\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]		\| [[File:Train result.png\|650px\|thumb\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]
	\|}		\|}
Line 282:		Line 282:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Test trueline avg.png\|250px\|thumb\|left\|alt=A\|'''Fig. 20. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Test trueline avg.png\|250px\|thumb\|left\|alt=A\|'''Fig. 20. ''' From up to dowm: R, G, B.( blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line)]]
	\| [[File:Test comare.png\|600px\|thumb\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]		\| [[File:Test comare.png\|600px\|thumb\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]
	\|}		\|}

imported>Student2018: /* CNN Architecture */

2018-12-17T00:44:32Z

CNN Architecture

← Older revision		Revision as of 00:44, 17 December 2018
Line 81:		Line 81:
	==== CNN Architecture ====		==== CNN Architecture ====

	Illustration of our CNN network architecture is given in Figure X. We used many convolution/deconvolution layers and relatively few pooling layers to increases the total amount of non-linearities in the network, which would help us learn complex mapping from IR to RGB. The activation function of each convolution layers is the ReLu function: <math> ReLu(x) = max(0,x) </math>, batch normalization is followed afterwards to further avoid overfitting.		Illustration of our CNN network architecture is given in Figure 8. We used many convolution/deconvolution layers and relatively few pooling layers to increases the total amount of non-linearities in the network, which would help us learn complex mapping from IR to RGB. The activation function of each convolution layers is the ReLu function: <math> ReLu(x) = max(0,x) </math>, batch normalization is followed afterwards to further avoid overfitting.

	[[File:CnnModel.png\|600px\|thumb\|center\|'''Fig. 8. ''' CNN Network Architecture Illustration]]		[[File:CnnModel.png\|600px\|thumb\|center\|'''Fig. 8. ''' CNN Network Architecture Illustration]]

imported>Student2018: /* CNN */

2018-12-17T00:40:45Z

CNN

← Older revision		Revision as of 00:40, 17 December 2018
Line 271:		Line 271:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Train trueline.png\|250px\|left\|alt=A\|'''Fig. 18. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Train trueline.png\|250px\|thumb\|left\|alt=A\|'''Fig. 18. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]
	\| [[File:Train result.png\|650px\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]		\| [[File:Train result.png\|650px\|thumb\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]
	\|}		\|}

Line 282:		Line 282:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Test trueline avg.png\|250px\|left\|alt=A\|'''Fig. 20. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Test trueline avg.png\|250px\|thumb\|left\|alt=A\|'''Fig. 20. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]
	\| [[File:Test comare.png\|600px\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]		\| [[File:Test comare.png\|600px\|thumb\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]
	\|}		\|}

imported>Student2018: /* Attempt 3 : RGB to IR Reverse Training */

2018-12-17T00:40:01Z

Attempt 3 : RGB to IR Reverse Training

← Older revision		Revision as of 00:40, 17 December 2018
Line 225:		Line 225:
	==== Attempt 3 : RGB to IR Reverse Training ====		==== Attempt 3 : RGB to IR Reverse Training ====

	[[File:RGBtoIR.png\| thumb\|~~400px~~\|alt=A\|'''Fig. 17. ''' RGB to IR testing (above) and training (below) ]]		[[File:RGBtoIR.png\| thumb\|500px\|alt=A\|'''Fig. 17. ''' RGB to IR testing (above) and training (below) ]]
	{\| class="L3 Model result" width="auto" style="margin-left: 0; margin-right: auto; border: none;"		{\| class="L3 Model result" width="auto" style="margin-left: 0; margin-right: auto; border: none;"
	\|-		\|-

imported>Student2018: /* Attempt 3 : RGB to IR Reverse Training */

2018-12-17T00:39:37Z

Attempt 3 : RGB to IR Reverse Training

← Older revision		Revision as of 00:39, 17 December 2018
Line 225:		Line 225:
	==== Attempt 3 : RGB to IR Reverse Training ====		==== Attempt 3 : RGB to IR Reverse Training ====

	[[File:RGBtoIR.png\| thumb\|~~500px~~\|alt=A\|'''Fig. 17. ''' RGB to IR testing (above) and training (below) ]]		[[File:RGBtoIR.png\| thumb\|400px\|alt=A\|'''Fig. 17. ''' RGB to IR testing (above) and training (below) ]]
	{\| class="L3 Model result" width="auto" style="margin-left: 0; margin-right: auto; border: none;"		{\| class="L3 Model result" width="auto" style="margin-left: 0; margin-right: auto; border: none;"
	\|-		\|-

imported>Student2018: /* Attempt 1 : Single Category Training */

2018-12-17T00:39:10Z

Attempt 1 : Single Category Training

← Older revision		Revision as of 00:39, 17 December 2018
Line 178:		Line 178:
	\|}		\|}

	[[File:Category.png\|500px\|center\|alt=A\|'''Fig. 12. ''' Categorized result]]		[[File:Category.png\|500px\|thumb\|center\|alt=A\|'''Fig. 12. ''' Categorized result]]

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"

imported>Student2018: /* L3 model result */

2018-12-17T00:38:38Z

L3 model result

← Older revision		Revision as of 00:38, 17 December 2018
Line 134:		Line 134:


	[[File:Rgb1.png\|450px\|center\|'''Fig. 9. ''' L3 rendered result]]		[[File:Rgb1.png\|450px\|thumb\|center\|'''Fig. 9. ''' L3 rendered result]]

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"

imported>Student2018: /* L3 model result */

2018-12-17T00:38:24Z

L3 model result

← Older revision		Revision as of 00:38, 17 December 2018
Line 133:		Line 133:
	\|}		\|}

	[[File:Rgb1.png\|~~500px~~\|center~~\|alt=A~~\|'''Fig. 9. ''' L3 rendered result]]
			[[File:Rgb1.png\|450px\|center\|'''Fig. 9. ''' L3 rendered result]]

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"

imported>Student2018: /* L3 Method */

2018-12-17T00:36:36Z

L3 Method

← Older revision		Revision as of 00:36, 17 December 2018
Line 19:		Line 19:

	The training step learns and stores the transformations used in rendering. Starting from scene spectral radiance data (left), two aligned representations are computed. Top: the sensor responses in a model camera system. Bottom: the target rendered image from scene radiance. The simulated sensor data at each pixel and its neighborhood (patch) is placed into one of several hundred pre-defined classes, based on properties such as the pixel type, response level, and local image contrast. For each class, the pairs of patch data and target output values are used to learn an affine transformation from the data and the target output value.		The training step learns and stores the transformations used in rendering. Starting from scene spectral radiance data (left), two aligned representations are computed. Top: the sensor responses in a model camera system. Bottom: the target rendered image from scene radiance. The simulated sensor data at each pixel and its neighborhood (patch) is placed into one of several hundred pre-defined classes, based on properties such as the pixel type, response level, and local image contrast. For each class, the pairs of patch data and target output values are used to learn an affine transformation from the data and the target output value.
	[[File:L32.png\|~~500px~~\|center~~\|alt=A~~\|'''Fig. 2. ''' L3 learning]]
			[[File:L32.png\|450px\|thumb\|center\|'''Fig. 2. ''' L3 learning]]

	====Dataset====		====Dataset====

imported>Student2018: /* L3 Method */

2018-12-17T00:35:56Z

L3 Method

← Older revision		Revision as of 00:35, 17 December 2018
Line 16:		Line 16:
	The rendering step adaptively selects from a stored table of affine transformations to convert the raw IR camera sensor data into RGB images, i.e. each pixel is classified by its local features into one of a few hundred classes (e.g. a red pixel, with high intensity, surrounded by a uniform field). Each class has a learned affine transform (stored in a table) that weights the sensor pixel values in order to calculate the rendered outputs. Hence the rendered output of each pixel is calculated as an affine transform of the pixel and its neighbors.		The rendering step adaptively selects from a stored table of affine transformations to convert the raw IR camera sensor data into RGB images, i.e. each pixel is classified by its local features into one of a few hundred classes (e.g. a red pixel, with high intensity, surrounded by a uniform field). Each class has a learned affine transform (stored in a table) that weights the sensor pixel values in order to calculate the rendered outputs. Hence the rendered output of each pixel is calculated as an affine transform of the pixel and its neighbors.

	[[File:L3.png\|~~500px~~\|thumb\|center\|'''Fig. 1. ''' L3 rendering]]		[[File:L3.png\|450px\|thumb\|center\|'''Fig. 1. ''' L3 rendering]]

	The training step learns and stores the transformations used in rendering. Starting from scene spectral radiance data (left), two aligned representations are computed. Top: the sensor responses in a model camera system. Bottom: the target rendered image from scene radiance. The simulated sensor data at each pixel and its neighborhood (patch) is placed into one of several hundred pre-defined classes, based on properties such as the pixel type, response level, and local image contrast. For each class, the pairs of patch data and target output values are used to learn an affine transformation from the data and the target output value.		The training step learns and stores the transformations used in rendering. Starting from scene spectral radiance data (left), two aligned representations are computed. Top: the sensor responses in a model camera system. Bottom: the target rendered image from scene radiance. The simulated sensor data at each pixel and its neighborhood (patch) is placed into one of several hundred pre-defined classes, based on properties such as the pixel type, response level, and local image contrast. For each class, the pairs of patch data and target output values are used to learn an affine transformation from the data and the target output value.

← Older revision		Revision as of 00:48, 17 December 2018
Line 271:		Line 271:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Train trueline.png\|250px\|thumb\|left\|alt=A\|'''Fig. 18. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Train trueline.png\|250px\|thumb\|left\|alt=A\|'''Fig. 18. ''' From up to dowm: R, G, B. (blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line)]]
	\| [[File:Train result.png\|650px\|thumb\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]		\| [[File:Train result.png\|650px\|thumb\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]
	\|}		\|}
Line 282:		Line 282:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Test trueline avg.png\|250px\|thumb\|left\|alt=A\|'''Fig. 20. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Test trueline avg.png\|250px\|thumb\|left\|alt=A\|'''Fig. 20. ''' From up to dowm: R, G, B.( blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line)]]
	\| [[File:Test comare.png\|600px\|thumb\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]		\| [[File:Test comare.png\|600px\|thumb\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]
	\|}		\|}

← Older revision		Revision as of 00:40, 17 December 2018
Line 271:		Line 271:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Train trueline.png\|250px\|left\|alt=A\|'''Fig. 18. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Train trueline.png\|250px\|thumb\|left\|alt=A\|'''Fig. 18. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]
	\| [[File:Train result.png\|650px\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]		\| [[File:Train result.png\|650px\|thumb\|upright\|alt=A\|'''Fig. 19. ''' Training set colorization]]
	\|}		\|}

Line 282:		Line 282:

	{\|style="margin: 0 auto;"		{\|style="margin: 0 auto;"
	\| [[File:Test trueline avg.png\|250px\|left\|alt=A\|'''Fig. 20. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]		\| [[File:Test trueline avg.png\|250px\|thumb\|left\|alt=A\|'''Fig. 20. ''' blue: Output RGB values v.s. Target RGB of Example Train image; orange: true line]]
	\| [[File:Test comare.png\|600px\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]		\| [[File:Test comare.png\|600px\|thumb\|upright\|alt=A\|'''Fig. 21. ''' Training set colorization]]
	\|}		\|}